New research from video editing company Kapwing shows that more than 20% of videos YouTube's algorithm serves to fresh accounts are low-quality AI-generated content. The study analyzed 15,000 of the platform's most popular channels and found that content farms producing nothing but automated videos have amassed over 63 billion views combined.
The methodology question
Kapwing's researchers created a new YouTube account and tracked the first 500 videos that appeared in its Shorts feed. Of those, 104 were AI slop and 165 qualified as "brainrot," a broader category that includes AI content alongside other low-effort videos designed to maximize attention. That's a third of a new user's initial experience.
The 21% figure for AI specifically raises some questions, though. Kapwing is a video editing company selling tools to creators. It has a business interest in demonstrating that human-made content is under threat. The study also focused on "trending" channels in each country, which might skew toward the kind of algorithmic gaming that AI content excels at.
Still, the numbers are hard to dismiss entirely. They surveyed the top 100 channels in every country on playboard.co and cross-referenced subscriber and revenue data from Social Blade.
Where the money goes
The most-viewed AI slop channel identified in the study is Bandar Apna Dost, an Indian channel with roughly 2.4 billion views. It features an anthropomorphic rhesus monkey and a muscular character modeled after the Incredible Hulk, fighting demons and traveling in what the Guardian describes as "a helicopter made of tomatoes." Kapwing estimates the channel could generate up to $4.25 million annually.
In the US, a Spanish-language channel called Cuentos Facinantes leads with 5.95 million subscribers, churning out Dragon Ball-themed videos. The channel was established in 2020, but its earliest currently available video dates to January 2025, which suggests either a content purge or a pivot to AI.
Spain's trending AI channels have accumulated 20.22 million subscribers despite having only eight such channels in its top 100. South Korea leads in total views, with 8.45 billion across 11 trending AI channels.
An underground economy
"There are these big swathes of people on Telegram, WhatsApp, Discord and message boards exchanging tips and ideas and selling courses about how to sort of make slop that will be engaging enough to earn money," journalist Max Read told the Guardian. Read, who has written extensively about AI slop for New York magazine, noted that creators often work in "niches." One recent trend: videos of pressure cookers exploding on stoves.
The people making this content tend to come from middle-income countries where YouTube ad revenue exceeds local wages. Read identified creators in Ukraine, India, Kenya, Nigeria, and Brazil as particularly active in these communities.
YouTube's awkward position
The platform finds itself caught between competing interests. YouTube CEO Neal Mohan has called generative AI the biggest change since the platform discovered people wanted to watch each other's videos. But advertisers don't want their brands adjacent to content that looks obviously automated.
YouTube updated its monetization policies in July 2025, renaming its "repetitious content" rule to "inauthentic content." The company clarified that AI tools are fine as long as they're used to enhance human creativity, not replace it. Whether that distinction holds up when enforcement relies on a combination of AI detection and human reviewers is another question.
A Guardian analysis from earlier in 2025 found that nearly 10% of the platform's fastest-growing channels were producing AI content exclusively. YouTube removed three channels and blocked ad revenue for two others after that investigation.
The reality is that the algorithm appears to know which videos are AI slop. Users who actively dislike and block these channels report that they stop appearing in their feeds. The fact that YouTube serves them to new accounts anyway suggests a choice about defaults, not a detection problem.
The Kapwing data is from October 2025. Given how fast this space moves, the numbers have probably shifted since then, though not necessarily in the direction you'd hope.
